• Aucun résultat trouvé

Quantum Shot Noise in Graphene

N/A
N/A
Protected

Academic year: 2021

Partager "Quantum Shot Noise in Graphene"

Copied!
228
0
0

Texte intégral

(1)Quantum Shot Noise in Graphene Andrey Mostovov. To cite this version: Andrey Mostovov. Quantum Shot Noise in Graphene. Other [cond-mat.other]. Université Pierre et Marie Curie - Paris VI, 2014. English. �NNT : 2014PA066061�. �tel-01023003�. HAL Id: tel-01023003 https://tel.archives-ouvertes.fr/tel-01023003 Submitted on 11 Jul 2014. HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés..

(2) ` THESE DE DOCTORAT ´ DE L’UNIVERSITE PIERRE ET MARIE CURIE Sp´ ecialit´ e : Physique ´ Ecole doctorale 107 : Physique en Ile-de-France. r´ ealis´ ee au. ´ Service de Physique de l’Etat Condens´ e, CEA Saclay Groupe Nanoelectronique pr´ esent´ ee par. Andrey MOSTOVOV pour obtenir le grade de : ´ PIERRE ET MARIE CURIE DOCTEUR DE L’UNIVERSITE. Sujet de la th` ese :. Quantum Shot Noise in Graphene pr´ esent´ ee et soutenue publiquement le 23 avril 2014 devant le jury compos´ e de :. M. M. M. M. M. M.. Dimitry RODITCHEV Massimo MACUCCI Fran¸ cois LEFLOCH Romain DANNEAU Denis Christian GLATTLI Preden ROULLEAU. Pr´ esident Rapporteur Rapporteur Examinateur Directeur de th` ese Invit´ e.

(3)

(4) Contents Introduction. 1. I Overview of Graphene Physics and Mesoscopic Transport 7 1 Physical Properties of Graphene 1.1 Crystal Structure of Graphene . . . . 1.2 Graphene Band Structure . . . . . . 1.2.1 Basic Principles . . . . . . . . 1.2.2 Band Structure Calculations . 1.2.3 Low-Energy Excitations . . . 1.3 Properties of Dirac Fermions . . . . . 1.3.1 Probabilty Current Density . 1.3.2 Zitterbewegung, Chirality and. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Klein Tunneling .. . . . . . . . .. 2 Transport in Mesoscopic Systems 2.1 Mesoscopic Scale . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Classical Discription of Transport . . . . . . . . . 2.1.2 Quantum Phase Coherent Systems . . . . . . . . 2.1.3 Effects of the Coherence on Transport Properties 2.2 Scattering Approach . . . . . . . . . . . . . . . . . . . . 2.2.1 Framework, Hypotheses, Formulation . . . . . . . 2.2.2 Scattering Approach at Work . . . . . . . . . . . 2.3 Quantum Hall Effect . . . . . . . . . . . . . . . . . . . . 2.3.1 Landau Quantization . . . . . . . . . . . . . . . . 2.3.2 Integer Quantum Hall Effect . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . .. . . . . . . . .. 9 9 11 11 13 16 18 19 19. . . . . . . . . . .. 23 24 24 26 27 30 30 32 36 38 40. 3 Mesoscopic Transport in Graphene 45 3.1 Conducance and Shot Noise in Graphene . . . . . . . . . . . . . . . 45 3.2 Quantum Hall Effect in Graphene . . . . . . . . . . . . . . . . . . . 49 i.

(5) ii. CONTENTS 3.3. II. Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . 52. Experimental Setup and Device. 59. 4 Measurement System Principle 4.1 Experimental Requirements and Techniques 4.1.1 Typical Scales . . . . . . . . . . . . . 4.1.2 Noise Measurement Techniques . . . 4.2 Technical Realisation . . . . . . . . . . . . . 4.2.1 Device Design . . . . . . . . . . . . . 4.2.2 Measurement System . . . . . . . . . 4.2.3 Cryogenic Amplification System . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. 61 61 61 63 65 66 67 70. 5 Device Fabrication 5.1 Methods . . . . . . . . . . . . . . . . . . 5.1.1 Obtaining Graphene . . . . . . . 5.1.2 Making Graphene Visible . . . . 5.1.3 Raman Spectroscopy of Graphene 5.1.4 Graphene Oxygen Plasma Etch . 5.2 Processes . . . . . . . . . . . . . . . . . 5.2.1 Wafers Preparation . . . . . . . . 5.2.2 Graphite Deposition . . . . . . . 5.2.3 Graphene Flakes Detection . . . . 5.2.4 Microcircuit Deposition . . . . . 5.2.5 Nano-Constriction in Graphene . 5.2.6 Device Test . . . . . . . . . . . . 5.2.7 Difficulties and Solutions . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. . . . . . . . . . . . . .. 75 76 76 80 82 85 85 85 87 89 90 94 95 96. III. . . . . . . . . . . . . .. . . . . . . . . . . . . .. Experimental Results. 6 Measurement System Calibration 6.1 Low Frequency Calibration . . . . 6.1.1 Lines Tuning . . . . . . . 6.1.2 Two-point Measurement . 6.1.3 Calibration . . . . . . . . 6.2 High Frequency Calibration . . . 6.2.1 Lines Tuning . . . . . . . 6.2.2 Current Measurement . .. 99 . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. 101 . 103 . 103 . 104 . 105 . 108 . 108 . 111.

(6) CONTENTS. iii. 7 Conductance Measurements at Zero Magnetic Field 7.1 Conductance at Zero Bias . . . . . . . . . . . . . . . . 7.1.1 Ballistic Regime Hypothesis . . . . . . . . . . . 7.1.2 Diffusive Regime Hypothesis . . . . . . . . . . . 7.1.3 Discussion . . . . . . . . . . . . . . . . . . . . . 7.1.4 Model for the “altered” Curves . . . . . . . . . 7.2 Conductance Spectroscopy . . . . . . . . . . . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. 119 119 120 122 125 128 130. 8 Conductance in the Magnetic Field 139 8.1 Measurement Principle . . . . . . . . . . . . . . . . . . . . . . . . . 139 8.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 142 9 Shot Noise in Graphene 9.1 Measurement Principle . . . . . . . . . . 9.2 Shot Noise in Presence of Joule Heating 9.2.1 Cooling by Electron Diffusion . . 9.2.2 Data Fit . . . . . . . . . . . . . . 9.2.3 Cooling by Phonon Emission . . . 9.3 Noise Power Fluctuations . . . . . . . . . 9.4 Results and Discussion . . . . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . .. Conclusion. . . . . . . .. 149 149 152 153 154 155 159 161 169. Appendices. 173. A Device Fabrication A.1 Recipes . . . . . . . . . . . . . . . . . . A.2 Introduction to Raman scattering . . . . A.3 Common Nano-fabrication Techniques . A.3.1 Microlithography Principle . . . . A.3.2 Optical and E-Beam lithography A.3.3 Thin Films Deposition . . . . . .. 173 . 173 . 173 . 175 . 175 . 176 . 177. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. B Measurement System 179 B.1 Cryogenic Inset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 B.2 Data Acquisition Module . . . . . . . . . . . . . . . . . . . . . . . . 179 B.3 RLC-Filter Pass-Band Calculation . . . . . . . . . . . . . . . . . . . 182 Bibliography. 185.

(7) iv. CONTENTS.

(8) R´ esum´ e en fran¸ cais R´ esum´ e de la th` ese Dix ans apr`es sa d´ecouverte, le graph`ene demeure une th´ematique tr`es en vogue en physique de la mati`ere condens´ee et en sciences des mat´eriaux. Cet unique mat´eriaux, un cristal v´eritablement bidimensionnel, qui a cout´e `a ses inventeurs A. Geim and K. Novoselov un Prix Nobel[? ] , est consid´er´e comme tr`es prometteur aussi bien de point de vue fondamental que pratique. La raison pour cela est la physique des ´electrons dans le graph`ene, qui sous certains conditions peuvent ˆetre vu comme des particules relativistes sans masse, d´ecrites par l’´equation de Dirac. Par cons´equent, le graph`ene pourrait servir comme un banc d’essai `a ´etat solid ´ pour des exp´eriences en Electrodynamique Quantique. Des effets m´esoscopiques sp´ecifiques comme l’Effet Hall Quantique non-conventionnel (relativiste) ou antilocalisation faible sont observ´es dans le graph`ene. D’autre part, ce mat´eriaux devrait permettre la fabrication des transistors extrˆemement rapides et des ´electrodes transparents, ainsi que trouver son application en photovolta¨ıque. Plusieurs autres applications sont envisag´ees.. Physique du Graph` ene Le graph`ene est plus connu sous forme de graphite, qui est tout simplement un tas des plans de graph`ene, tenus par les forces de van der Waals. Les atomes de carbone dans le graph`ene sont plac´es sur un r´eseau hexagonal (ou en nid d’abeilles). Ce r´eseau ne constitue pas un r´eseau de Bravais, faute de sa sym´etrie et la maille ´el´ementaire contient deux atomes. Par cons´equent, il est d’usage de distinguer deux sous-r´eseaux d’atomes A et B (voir fig. 1.a). Le r´eseau de Bravais dans ce cas est de type triangulaire avec des vecteurs de base a1 =. √. 3aex. and. a2 = v. √. √ � 3a � ex + 3ey . 2. (1).

(9) vi. CONTENTS. . . .   . . . . . . .

(10) . . 

(11) . . . Figure 1: (a) R´eseau hexagonal (ou en nid d’abeilles) du graph`ene avec deux sous-r´eseaux en couleurs diff´erentes, distance interatomique a = 0.142 nm. (b) La premi`ere zone de Brillouin et les points K et K � (deux seulement sont in´equivalents). Le r´eseau r´eciproque est engendr´e par les vecteurs suivants a1∗. �. 2π ey =√ ex − √ 3a 3. �. and. a2∗ =. 4π ey , 3a. (2). et la premi`ere zone de Brillouin est repr´esent´e sur fig. 1.b. Deux points particuliers situ´es sur les coins de la premi`ere zone de Brillouin seront importants dans la suite. Nous les notons K et K � , ils sont donn´es par les vecteurs 4π ±K = ± √ ex . 3 3a. (3). Dans l’approximation de liaisons fortes on peut chercher les fonctions d’ondes propres du syst`eme sous forme de combinaison lin´eaire des fonctions de Bloch de deux sous-r´eseaux A et B: B ψk (r) = ak ϕA k (r) + bk ϕk (r). (4). Les valeurs propres de l’´equation de Schr¨odinger Hψk = �k ψk seront donn´ees par l’´equation s´eculaire1 . 1. . −εk t · γk  det  =0 t · γk† −εk. (5). Ici nous somme dans une approximation qui consiste `a restreindre le couplage aux premiers voisins uniquement et n´egliger le recouvrement entre les orbitaux atomiques..

(12) CONTENTS. vii. t ´etant l’int´egrale de transfert (t ≈ −3 eV) et γk ≡ 1 + e−ık·a2 + e−ık·(a2 −a1 ) est un pr´efacteur de phases des premier voisins, venant du fait que chaque atome a trois premiers voisins, dont deux n’appartiennent pas au mˆeme nœuds du r´eseau de Bravais que lui-mˆeme. Cela donne deux solutions: ε± k = ±t|γk | =. � � 3 � � ±t�3 + 2 cos (k · a. i). (6). i=1. Les solutions positives et n´egatives correspondent aux bandes de Conduction et de Valence respectivement. Les deux bandes se touchent en deux points de la premi`ere zone de Brillouin d´ej`a mentionn´e: K et K � . En particulier, la relation de dispersion est lin´earisables au voisinage de ses deux points. On utilise la notation suivante: k = ±K + q o` u le vecteur q est tel que |q| � |K|. Cela r´esulte en Hamiltonien effectif lin´earis´e autour du point K (pour K � il faut inverser le signe): � = �v qσ H q F. (7). ε± q = ±�vF |q|,. (8). avec vF ≡ − 3ta ≈ 1 × 106 m s−1 — vitesse de Fermi and σ ≡ (σ x , σ y ) — vecteur 2� de matrices de Pauli matrices. Mis sous cette forme, ce Hamiltonien est souvent appel´e le Hamiltonien de Dirac par analogie avec la physique des particules. La relation de dispersion lin´earis´ee est finalement donn´ee par:. aussi bien pour la lin´earisation autour du point K, que autour du point K � , ce qui r´esulte en d´eg´en´erescence suppl´ementaire dite de vall´ee. Il est courant, en vue de la forme du Hamiltonien (7), d’employer la notation vectorielle ou d’un spinor (´egalement par similitude avec la notation en physique des particules) pour d´ecrire les ´etats du syst`eme: Ψ± k. �. a± k = ± bk. �. (9). o` u les composantes de ce vecteur sont les facteurs de phase avec lesquelles les deux sous r´eseaux contribuent `a l’´etat donn´e. En particulier, les vecteurs propres du nouveau Hamiltonien s’´ecrivent Ψ± k. �. �. 1 1 =√ , ıφ 2 ±e k. �. Imγk avec φk = arctan Reγk. �. (10). Cela n’est pas superflu de souligner ici que les caract´eristiques extraordinaires de ce mat´eriau sont surtout d´etermin´ees par ces propri´et´es physiques singuli`eres..

(13) viii. CONTENTS. Transport M´ esoscopique dans le Graph` ene Dans ce travail, le graph`ene est examin´e du point de vu de la physique m´esoscopique, en particulier le mesures de la conductance et du bruit sont r´ealis´ees. Ce champ de physique de la mati`ere condens´ee s’int´eresse aux syst`emes dont la taille caract´eristique appartient `a l’´echelle“interm´ediaire” entre l’´echelle microscopique et l’´echelle macroscopique. La limite inf´erieure de cette ´echelle peut ˆetre donn´ee par la taille d’un atome, alors que la limit´e sup´erieur est d´etermin´ee par la longueur de coh´erence de phase lφ , la distance maximale sur laquelle la coh´erence quantique de phase des ´etats est pr´eserv´ee. Sous ces conditions les effets quantiques tels que les fluctuations quantiques, les interf´erences quantiques et parfois ´egalement les interactions commencent a jouer un rˆole important. Afin de briser la coh´erence de phase une diffusion in´elastique est n´ecessaire et `a basse temp´erature, o` u le couplage ´electron-phonon est faible, celle-ci est due aux interactions ´electron-´electron. En dessous de lφ on peut ´egalement distinguer deux r´egimes diff´erents. Si en traversant l’´echantillon, les ´electrons subissent un grand nombre de collisions ´elastiques, c’est un r´egime dit diffusif. Alors que, si en traversant l’´echantillon ils ne subissent aucun choc, c’est un r´egime dit balistique. La longueur caract´eristique de transition entre ces deux r´egimes s’appelle le libre parcours moyen, le . L’approche le plus rependu de traiter les probl`emes `a l’´echelle m´esoscopique est celle de la th´eorie de la diffusion propos´e par R. Landauer[91, 93, 94, 19, 92] et d´evelopp´ee ensuite par M. B¨ uttiker[19, 20, 92, 16, 17, 18] , Y. Imry[72, 19, 20, 73, 141, 74] et d’autres personnalit´es. L’id´ee de cette th´eorie consiste a relier les propri´et´es de transport du syst`eme ´etudi´e (e.g. conductance, les fluctuations du courant) `a ses propri´et´es de diffusion, consid´er´ees d’ˆetre connues `a partir des calculs de la m´ecanique quantique. Par exemple analysons le cas d’un conducteur `a deux terminaux (le gauche et le droit), connect´e `a deux r´eservoirs d’´electrons (voir fig. 2). On peut distinguer deux types d’´etats aussi bien `a gauche qu’`a droite: les ´etats incidents sur le conducteur et les ´etats quittant le conducteur vers l’un ou l’autre r´eservoir. Le bilan de tous ses ´etats va d´eterminer les propri´et´es de transport du syst`eme ´etudi´e. La th´eorie de la diffusion propose donc de calculer la matrice dite de la diffusion reliant les ´etats incidents aux ´etats quittant le conducteur `a gauche et `a droite. Les ´el´ements de cette matrice permettent de parvenir jusqu’aux coefficients de la transmission ou de la diffusion de chaque mode de propagation, Dn et Rn = 1 − Dn , n ´etant l’indice de mode (´evidemment Dn ≤ 1). Mentionnons deux principaux r´esultats de cette th´eorie. Tout d’abord la conductance d’un conducteur coh´erent est donn´ee par la formule G= la quantit´e. 2e2 , h. 2e2 � Dn h n. (11). appel´ee le quantum de conductance, d´efinit la conductance maxi-.

(14) CONTENTS. ix. [14] Blanter and B¨ uttiker (1999). Figure 2: Exemple du probl`eme d’un conducteur `a deux terminaux dans la th´eorie de la diffusion. Adopt´e de [14]. male que peux avoir un mode de transmission. Ce r´esultat est d’une tr`es grande port´ee, car il montre que mˆeme en absence de toute diffusion la conductance n’est pas infini, ce qu’un mod`ele classique n’est pas capable de pr´edire. En effet, ce fait est une cons´equence `a la fois du principe de Pauli et du prince d’incertitude d’Heisenberg. Le premier limite le nombre d’´electrons par ´etat `a deux (spin up et spin down), alors que le dernier d´efini l’espacement sur l’axe temporel des paquets d’ondes d’une tranche d’´energie donn´ee (´egale `a µL −µR ) et ainsi limite la fr´equence d’injection des paquets dans le conducteur. Le deuxi`eme r´esultat concerne le bruit de partition. La densit´e spectrale de “pur” bruit de partition (i.e. `a temp´erature nul) s’´ecrit SI = 2eI. �. n. Dn (1 − Dn ) � n Dn. (12). et si le bruit thermique est pris en compte, on obtient �. eV 2e2 � 2 2e2 � Dn kB T + 2e Dn (1 − Dn )V coth SI = 4 h n h n 2kB T. �. (13). On introduit alors une quantit´e qu’on appel le facteur de Fano: F=. �. n. Dn (1 − Dn ) � n Dn. (14). La conductance et le facteur de Fano sont les caract´eristiques principales d’un conducteur m´esoscopique. Prenons quelques exemples type. Le facteur de Fano d’un conducteur balistique est ´evidemment nul, tandis que pour une jonction tunnel o` u tous les coefficients de transmission sont extrˆemement faibles, il vaut un. Le cas d’un conducteur diffusif est plus compliqu´e, car il n’y est pas possible de calculer pr´ecis´ement les coefficients de transmission Dn . Seulement la distribution de probabilit´es pour ces coefficients de prendre une valeur est calculable. Avec ce r´esultat le facteur de Fano ´egal `a 1/3 a ´et´e d´eduit..

(15) x. CONTENTS. [148] Tworzydlo et al. (2006). Figure 3: D´ependance en ´energie de la conductivit´e σ (a) et du facteur de Fano F (b) pour un rapport d’aspects W/L = 5. Le minimum de la conductivit´e au point de Dirac correspond au maximum du facteur de Fano. Adopt´e de [148]. Revenons `a pr´esent au graph`ene. Les propri´et´e du transport coh´erent d’un ruban de graphene id´eal ont ´et´e calcul´ees par Tworzydlo et al. de fa¸con analytique. Les coefficients de la transmission sont donn´es par la formule suivante Dn =. � �2 � � kn � � � � , � kn cos (kn L) + ı(µ/�vF ) sin (kn L) �. (15). o` u µ est le potentiel�´electrochimique du conducteur, qn — le moment transversal ` l’aide de cette formule, la conductance (ou du mode n et kn = (µ/�vF )2 − qn2 . A la conductivit´e σ ≡ G × L/W ) et le facteur de Fano sont calcul´es. Dans la limite W/L → ∞ (d´ej`a satisfaite pour W/L � 4) on conclue, que ses quantit´es au point de neutralit´e prennent des valeur universelles: 4e2 σ→ , πh. F → 1/3,. lors que µ = 0. (16). Les r´esultats plus complets son pr´esent´es sur fig. 3. La conductivit´e atteint donc son minimum au mˆeme temps que le facteur de Fano atteint son maximum et cela au point de neutralit´e (aussi appel´e le point de Dirac). Il est remarquable que dans un ruban de graph`ene id´eal et donc balistique le facteur de Fano prenne au point de neutralit´e la valeur 1/3, qui est celle d’un conducteur diffusif. Ce fait a suscit´e beaucoup d’int´erˆet. D’autre part, pour le graph`ene diffusif il n’existe pas de mod`ele analytique et il n’y pas d’unanimit´e sur la valeur exacte du facteur de Fano ainsi que de sa d´ependance en potentiel ´electrochimique et en degr´e de d´esordre. Citons quelque r´esultats num´eriques. San-Jose et al. examinent le probl`eme dans le cadre du.

(16) CONTENTS. xi. [42] DiCarlo et al. (2008) [35] Danneau et al. (2008). Figure 4: R´esultats pr´ec´edents. Mesure par DiCarlo et al. de la r´esistance (a) et du facteur de Fano (b). Adopt´e de [42]. (c) et (d) — mesures des mˆemes quantit´es par Danneau et al., adopt´e de [35] formalisme des matrices de transfert en consid´erant un potentiel de d´esordre lisse uni- et bi-dimensionnel. Ils obtiennent le valeurs F1D = 0.243 et F2D = 0.295 dans ces deux cas respectifs. Lewenkopf et al. ont utilis´e la m´ethode des fonctions de Green r´ecursives pour le d´esordre `a longue port´ee. Ils concluent pour un niveau de d´esordre mod´er´e, que le facteur de Fano est situ´e autour de 0.25−0.29 (en fonction du degr´e de d´esordre), mais ´egalement pr´edisent l’apparition d’un pic au point de neutralit´e. Logoteta et al. ´etudie le facteur de Fano en fonction du potentiel ´electrochimique en r´esolvant l’´equation de Dirac par la m´ethode des fonctions enveloppes et de Fourier[103] . Pour les rapports d’aspects et les concentrations d’impuret´es qui nous int´eresseront, ils trouvent que F oscille autour de 0.25 et atteint au point de neutralit´e 0.35. Finalement, regardons des travaux d’exp´erimentateurs `a ce sujet, pr´esent´es en partie sur la fig. 4. Seulement deux ´etudes du bruit dans le graph`ene ont ´et´e men´ees pr´ec´edemment. La premi`ere, a ´et´e r´ealis´ee `a l’Universit´e de Harvard par DiCarlo et al. Leurs ´echantillons se sont montr´es diffusifs et ils ont observ´e.

(17) xii. CONTENTS. le facteur de Fano situ´e pr`es de 0.35 et pratiquement ind´ependant du potentiel ´electrochimique. Quant `a la conductivit´e au point de neutralit´e, celle-ci `a large2 ment d´epass´e la valeur 4e . La deuxi`eme ´etude r´ealis´ee par Danneau et al. `a πh l’Universit´e de Helsinki montre sur un ´echantillon que les valeurs universelles de la conductivit´e et du facteur de Fano sont atteint, alors que leurs mesures sur les autres ´echantillons sont moins concluantes. Il reste donc bien des questions ouvertes atour de ce sujet.. Pr´ esentation de l’approche exp´ erimentale Nous pr´esentons ici une troisi`eme exp´erience de bruit quantique dans le graph`ene, compl´etant les deux travaux ci-dessus. Dans notre exp´erience nous avons eu recours `a un nombre de techniques de mesures de bruit pour ainsi obtenir des conditions exp´erimentales les plus favorables et d´epasser certaines limitations des travaux pr´ec´edents. Notamment dans notre exp´erience nous appliquons la m´ethode de corr´elations crois´ees qui, au prix d’un doublement du nombre des lignes de d´etection, permet de r´eduire significativement l’impacte du syst`eme de d´etection sur le bruit d´etect´e. De plus, nous faisons une mesure quatre points, qui a des avantages ´evidents pour une mesure de conductance mais ´egalement pour celle du bruit, car elle exclut la source parasite de bruit thermique que sont les r´esistances de contacte. Utilisation des amplificateurs cryog´eniques bas bruit (faits maison) plac´es `a proximit´e de l’´echantillon et coupl´es aux circuits r´esonants (fr´equence de r´esonance � 3.3 MHz) rendent la d´etection plus efficace et rapide et la contribution du bruit en 1/f n´egligeable. Finalement la chute du potentiel est retrouv´ee grˆace `a deux amplificateurs diff´erentiels (situ´es `a temp´erature ambiante), les signaux sont num´eris´es par une carte rapide d’acquisition (´echantillonnage `a 10 MHz) et la puissance spectrale est calcul´e par l’ordinateur `a l’aide d’un module de transform´e de Fourier rapide. Le principe de notre syst`eme exp´erimental est d´ecrit dans la sec. 4.2. Notre ´echantillon comporte ´egalement des avantages (voir subsec. 4.2.1). Premi`erement, il a une g´eom´etrie sp´eciale: nous ´etudions le transport non pas `a travers toute la couche de graph`ene, mais seulement `a travers une ´etroite constriction que poss`ede la couche. La constrictions ´etant beaucoup plus ´etroite que le reste de la couche, ces propri´et´es vont dominer le transport `a travers l’´echantillon. Les probl`emes de mauvais contacts entre les r´eservoirs d’´electrons et le conducteur (i.e. la r´egion ´etudi´ee) seront ainsi ´evit´es, car les trois font partie de la mˆeme couche de graph`ene. Finalement, afin d’´echapper aux capacit´es parasites par l’interm´ediaire du substrat nous avons d´epos´e le graph`ene sur un substrat de silicium non-dop´e et utilisons des grilles lat´erales `a la place de la grille arri`ere. Toutes les ´etapes de fabrication de l’´echantillon ont ´et´e effectu´e par l’auteur et uniquement dans le laboratoire d’origine, elles sont expos´ees dans le Chapitre 5. Sch´ema du cir-.

(18) CONTENTS. xiii .  .  .   . 

(19).  . . . Figure 5: Dessin du dispositif `a base de graph`ene utilis´e dans nos exp´eriences: 1 — deux ´electrodes d’injection du courant, 2 — quatre sondes de tension, 3 — grilles lat´erales (d´econnect´ees ´electriquement du graph`ene), 4 — nano-constriction, jouant le rˆole d’un conducteur coh´erent (proportions ne sont pas conserv´ees pour meilleure visibilit´e). L’ensemble est situ´e sur un substrat de silicium isolant (en violet), contenant une couche de SiO2 (in bleu claire). cuit de mesure avec un dessin de l’´echantillon au centre: six contacts ohmiques sont color´es en jaune et les grilles lat´erales en vert, la constriction est montr´ee explicitement et not´ee Rs . Quatre lignes de mesure de tension identiques sont connect´ees `a l’´echantillon, chacune contient un circuit r´esonant et un amplificateur cryog´enique. A temp´erature ambiante elles sont connect´ees `a deux amplificateurs diff´erentiels NF SA-420F5, ce qui r´esulte au final en deux canaux de sortie (ch0 et ch1), qui `a leur tour sont branch´es sur les entr´ees des diff´erents instruments de mesure, y compris la carte rapide acquisition (aussi dessin´ee). Une grande partie du circuit (d´elimit´ee sur le sch´ema par une ligne de tirets) est situ´ee dans une canne frigorifique, les temp´eratures exp´erimentales possibles sont indiqu´ees. En revanche, le prix `a payer pour les b´en´efices remarquables que repr´esente l’architecture de notre syst`eme exp´erimental est une fabrication des ´echantillons extrˆemement laborieuse (voir Chapitre 5), mais aussi une calibration du syst`eme exp´erimental difficile et tr`es subtile, due au fait que les amplificateurs cryog´eniques employ´es ont un gain ajustable et apriori inconnu. Par cons´equent, nous devions nous assurer d’abord que les quatre lignes de mesure de tension, que nous utilisons pour la d´etection du bruit, ont pr´ecis´ement le mˆeme gain chacune et ensuite ` la proc´edure de calibration est consacr´e le Chapitre 6 de ce d´eterminer ce gain. A manuscrit.. Pr´ esentation des r´ esultats Avant de passer `a la mesure du bruit nous avons men´e quelques mesures pr´eliminaires. La conductance de notre ´echantillon en fonction de la tension de grille a ´et´e d´etermin´ee (voir Chapitre 7). A la base, notre ´echantillon est dop´e en.

(20) xiv. CONTENTS. RS (kΩ). trous, mais grˆace aux grilles lat´erales nous pouvons largement d´epasser le point de neutralit´e et doper notre ´echantillon en ´electrons. La comparaison des r´esultats de cette mesure avec les pr´edictions th´eoriques pour le r´egime balistique et celui diffusif nous permet de confirmer le caract`ere diffusif de notre ´echantillon (voir subsec. 7.1.1 et subsec. 7.1.2). Nous avons pu ´egalement estimer la mobilit´e, le libre parcours moyen, la densit´e d’impuret´es, la longueur de coh´erence et certains autres caract´eristiques de notre ´echantillon, ce que nous d´ecrivons dans la subsec. 7.1.3.. 10 9 8 7 6 5 4 3 2 1 0 -20. Experimental signal Simulations. 0. 20. 40. 60. 80 100 120. Gate voltage (V) Figure 7: La r´esistance de la constriction en fonction de la tension de grille (vert et bleu). Les r´esultats d’ajustement des param`etres en consid´erant le mod`ele de transport diffusif d´ecrit dans le texte (en violet) et une am´elioration de ce mod`ele (en rouge). En outre, nous avons sond´e les propri´et´e non-lin´eaires de notre ´echantillon par une mesure de conductance diff´erentiel en fonction de la tension DC appliqu´ee (voir sec. 7.2). Cela nous sert entre autre `a relier lors de la mesure du bruit le courant DC que nous injectons dans l’´echantillon et la tension qui se forme au bord de la constriction. Une mesure dans le r´egime d’Effet Hall Quantique a ´et´e aussi effectu´ee, ce que nous d´ecrivons dans le Chapitre 8. Celle-ci nous a notamment permis de connaˆıtre la densit´e de porteurs de charge en fonction de la tension de grille appliqu´ee. Finalement, nous avons proc´ed´e aux mesures du bruit. Les exp´eriences pr´ec´edentes nous ont servi de correctement interpr´eter nos mesures (voir sec. 9.1). A partir des r´esultats de celles-ci nous avons conclu, que la temp´erature des ´electrons dans la constriction d´epasse la temp´erature du bain d’h´elium. Nous avons tent´e.

(21) CONTENTS. (a). 25. R(kΩ) B(T ). xv. 15. 8.6kΩ. 5 R(kΩ) 7.5 LL0 LL-1 LL1 25.0 ν =-6 ν=6 ν =-2 ν=2 7.0 22.5 20.0 6.5 17.5 15.0 6.0 12.5 10.0 5.5 7.5 5.0 5.0 4.5 -55 -45 -35 -25 -15 -5 5 15 25 35 45 55. ∆VG(V ). Figure 8: Un graphique couleur de la r´esistance mesur´ee avec la combinaison crois´ee des sondes de tension (voir le texte) en fonction de ∆VG = VG − VD et du champ magn´etique B. Les courbes du haut montre la variation de la r´esistance au champ magn´etique maximal B = 7.565 T (courbe rouge) et de la r´esistance moyenn´ee sur les lignes (radiales) de facteur de remplissage constant (courbe verte). d’expliquer ce fait par la dissipation dans un des contacts d’injection du courant, qui s’est av´er´e de tr`es haute r´esistance. Initialement le mod`ele de propagation de la chaleur par la diffusion ´electronique seule a ´et´e envisag´e (conductance thermique donn´ee par la loi Wiedemann-Franz, voir subsec. 9.2.1). En utilisant ce mod`ele la d´ependance de la temp´erature ´electronique du courant inject´e a ´et´e d´etermin´ee pour l’utiliser dans le mod`ele du bruit (repr´esent´e `a temp´erature constante par l’´equation (13), voir subsec. 9.2.2). De l`a nous avons extrait le facteur de Fano et une raisonnable temp´erature (proche de celle du bain d’h´elium) de l’´etage o` u l’´echantillon est thermiquement ancr´e. En revanche, le troisi`eme param`etre ajustable extrait correspondait a une r´esistance du graph`ene plus faible que ce que nous avons estim´e. Par cons´equent, nous avons ´evoqu´e le m´ecanisme de refroidissement d’´electrons par le couplage aux phonons d´ecrit dans la subsec. 9.2.3. Nous avons calcul´e la distance apr`es laquelle la temp´erature ´electronique sera peu diff´erente de la temp´erature du r´eseau cristallin. Cette distance caract´eristique est comparable `a la distance entre la source de chaleur et la constriction, ce qui confirme l’efficacit´e de ce deuxi`eme m´ecanisme..

(22) CONTENTS. 1,0. 3. r.m.s. d(δSII)/dVds (α⋅8e /h). xvi. 0,8. e. 0,6 0,4. VG = 40V VG = 50V. 0,2 0,0. VG = 67.5V -5 -4 -3 -2 -1 0. 1. 2. 3. 4. 5. VDS(mV) ´ Figure 9: Etude des fluctuations de la puissance spectrale du bruit � d’apr`es la 46 . r´elation (17). La valeur th´eorique correspond `a 1 dans ces unit´es, α = 2835 Dans la sec. 9.3 nous nous sommes ´egalement int´eress´e aux irr´egularit´es des courbes de bruit obtenues. Ces irr´egularit´es sont les fluctuations de la puissance spectrale du bruit, qui ont le mˆeme origine que les fluctuations universelles de la conductance `a savoir les interf´erences complexes entre les diff´erents chemins de propagation dans un conducteur diffusif, qui d´ependent de l’´energie de la particule inject´ee et vont ainsi impacter les coefficients de la transmission. Ce ph´enom`ene a ´et´e examin´e de fa¸con th´eorique dans des gazes bidimensionnels d’´electrons conventionnels par de Jong and Beenakker, qui ont propos´e l’expression suivante pour les fluctuations de la densit´e spectrale: 2e2 r.m.s. δSII = 2e|Vds | h. �. 46 2835. (17). Applicabilit´e de cette formule dans le cas du graph`ene a ´et´e v´erifi´ee. Nos r´esultats sont mont´es sur la fig. 9 et expos´es de fa¸con plus d´etall´ee dans la sec. 9.3. Nous concluons que (17) est en effet applicable `a un coefficient num´erique pr`es. Nos mesures nous fournissent la d´ependance du facteur de Fano de la tension de grille repr´esent´ee sur la fig. 10. La variation avec la tension de grille est faible, le facteur de Fano reste proche de ≈ 0.25. En revanche, les fluctuations du facteur de Fano sont bien prononc´ees. Comme il a d´ej`a ´et´e not´e, notre ´echantillon est diffusif, le comportement du facteur de Fano n’est pas d´ecrit par le mod`ele propos´e par Tworzydlo et al. pour.

(23) CONTENTS. xvii. Fano Factor. 0,6. 0,4. 0,2. 0,0 30. 40. 50. 60. 70. 80. 90. VG(V) Figure 10: Facteur de Fano en fonction de la tension de grille, extrait des donn´ees exp´erimentales par l’ajustement des trois param`etres, la ligne horizontale en pointill´es correspond `a la valeur 1/3. Le point de Dirac se trouve entre 58 et 61 V. un ruban id´eal (balistique) de graph`ene. En revanche, nos r´esultats pr´esentent une forte similitude avec ceux de Danneau et al. pour leur ´echantillon diffusif (´echantillon E). Quant `a DiCarlo et al., il ont mesur´e des valeurs ≈ 0.35, de 45% sup´erieur par rapport aux nˆotres et le facteur de Fano dans leur mesure ne varie pas du tout avec la tension de grille. Nous attribuons cette disparit´e entre les deux r´esultats aux diff´erences dans le degr´e du d´esordre entre deux ´echantillons. L’accord avec les r´esultats th´eoriques, pr´esent´es au d´ebut de ce r´esum´e (voir les travaux de San-Jose et al., Lewenkopf et al., Logoteta et al.) est ´egalement bon. Notamment, le pic au point de neutralit´e, pr´edit par Lewenkopf et al. et par Logoteta et al. est clairement visible sur la fig. 10. Une analyse plus cons´equent de nos r´esultat se trouve dans la sec. 9.4..

(24) xviii. CONTENTS.

(25) Introduction Ten years after its discovery, graphene remains the hottest topic of the condensed matter physics and the materials science. This unique, truly two-dimensional carbon based crystal “costed” their inventors A. Geim and K. Novoselov a Nobel Prize[? ] . Today, new, more “industrializable” deposition techniques are ready to replace the micromechanical cleavage, once allowed to isolate the first graphene layer[114] . These technique (CVD, SiC sublimation, MBE), already employed in the laboratories, can in a not so far future bring graphene into the everyday life devices. So, let us see, what graphene is exactly about. The carbon atoms in graphene are arranged in a hexagonal or honeycomb lattice as shown in fig. 11.a. Particular symmetries of this lattice together with the confinement in two-dimensions result in what became the hallmark of graphene: to some extent, electrons in graphene are considered to have zero effective mass and obey Dirac Equation, described by matrix Hamiltonian � � 0 qx − ıqy Hq = �vF (18) qx − ıqy 0 where q is the wave vector with respect to one of two very singular points of graphene Brillouin Zone, called Dirac points and denoted K and K � . This equation leads to a linear dispersion relation around each of these two points, known as Dirac cones or valleys, shown in fig. 11.a, yielding additional two-fold valley degeneracy. The above-mentioned facts determine most of the singular physical properties of this unique material. But, why exactly is this material so fascinating? For the fundamental scientists, graphene appeared attractive promptly after revealing its potential to turn into a solid-state test bed for the Quantum Electrodynamics (QED). Low-energy electronic excitations in graphene, being massless chiral Dirac fermions with an effective speed of light ≈ c/300[115] , hence mimic their QED counterparts (like neutrino for instance) and could probably allow to observe in an nanometre-size 2D crystal certain High Energy Physics phenomena, for which usually atom-smasher-scale facilities are required. In particular, some relativistic effects, theoretically predicted, but experimentally not yet confirmed, such as Klein paradox[86] (transmission of particles through a classically forbidden 1.

(26) 2. CONTENTS. [158] Williams (2009). Figure 11: (a) Graphene honeycomb lattice with two sublattices A and B highlighted with different colours. Interatomic distance a = 0.142 nm. (b) Dirac cone. From [158] region with probability close to one) or Zitterbewegung[133] (jittery motion of a particle, due to interference between components of its wavefunction belonging to positive (electron) and negative (positron) energy states), are likely to manifest themselves in graphene[82, 81] . On the other hand, several singularities of graphene physics present an interest for a condensed matter physicists, regardless of any QED considerations. To name only few, unconventional Quantum Hall Effect, observed even at room temperature thanks to the the large cyclotron energies of electrons in graphene[116] , Quantum Hall Isospin Ferromagnetism, due to its fourfold spin–valley degeneracy and the strong Coulomb interactions[166] or 2π Berry phase-related weak anti-localization effect enhancing the classical Drude conductivity[161] . In addition, pronounced ambipolar electric field effect and predicted high room-temperature mobilities even up to 200 000 cm2 V−1 s−1[27] (such mobilities, but at 5 K and in suspended samples were reported by Bolotin et al.) make of graphene a very appealing material for experimentalist, but also for microelectronics engineers. Graphene-based high-speed electronics is among the most promising applications of this material: implementation of a room-temperature ballistic transistor, operating at THz frequencies would be a ground breaking achievement of the microelectronics. Recently reported 427 GHz graphene transistor[31] and new graphene-adapted logic circuits architecture[98] together with latest numerical results[174] , make believe that the idea of large-scale manufacturing of such devices is not just hypothetical. Another, more trivial application of graphene in the field of microelectronics are flexible transparent electrodes. Microelectronics is not the only sphere of graphene application. Other poten-.

(27) CONTENTS. 3. [148] Tworzydlo et al. (2006). Figure 12: Fermi energy dependence of the conductivity σ (a) and the Fano factor F (b) at fixed aspect ratio W/L = 5. The conductivity minimum at the Dirac point corresponds to maximal Fano factor. From [148].. tial application is the photovoltaics. According to the recent study by Tielrooij et al., in graphene the internal quantum efficiency is enhanced through “hot-carrier multiplication” cascade-type process. In a nutshell, photoexcited carriers transfer their excess energy to other charge carriers, creating additional electron-hole pairs, instead of heating the lattice as it is often the case in other materials. In addition, by placing an appropriate plasmonic nanostructures near graphene p-n junction, the efficiency can be further enhanced by order of magnitude[46] . Besides, graphene is a very hopeful candidate for the development of Terahertz domain detectors[79, 154] , which is a challenging task of today’s photonics. Finally, it is also worth citing several other applications as chemical and bio sensors[134] , nanoelectromechanical devices[69] , data storage devices[77] , supercapacitors[168, 118] and very efficient membranes[113] , the list is not exhaustive. In the present work we approach graphene from the position of the mesoscopic physics. This field of condensed matter physics, emerged in early 80s, study low-dimensional systems of micrometer size in which quantum effects such as fluctuations and interferences are observable (very low temperatures are also generally required). It is obvious, that graphene, being a standalone two-dimensional system, presents an extraordinary opportunity for mesoscopic physicists. Most of emblematic mesoscopic physics phenomena (e.g. Integer and Fractional Quantum Hall effects, Shubnikov–de Haas and Aharonov-Bohm effects and many others, see refs. [173, 41, 166, 164, 145, 160, 45, 147, 29, 10, 122, 162, 89]) were tested in this system..

(28) 4. CONTENTS. More specific to graphene problem of minimal conductivity and Fano factor2 at Dirac point was addressed by Tworzydlo et al. in the case of short and large ideal (ballistic) graphene nano-ribbons. Minimal conductivity is an interesting problem, since on one hand graphene is a gapless semi-metal and isolator state is not attainable in it. On the other hand, precisely at the Dirac point the carrier density is zero, so one would expect the conductivity going to zero as well. The conclusions of Tworzydlo et al. were that minimum of conductivity is indeed reached at the 2 Dirac point, yet its value is not zero but rather 4e . As for the Fano factor, it πh was shown that in contrast to what happens in conventional ballistic systems, at the Dirac point this quantity is not zero but equals 1/3 (same value is observed in diffusive systems), tending to zero only far from it. Both results are shown in fig. 12. Two experimentalists groups tried to challenge these theoretical predictions. First group from Harvard University measured shot noise in graphene samples of several aspect ratios and lengths as well as in a graphene p-n junction, see ref. [42]. However, these samples turned out to be diffusive, i.e. having mean free path (≈ 40 nm) much shorter than their length. As a consequence, the observed behaviour of Fano factor with injection energy was found to be the usual one for diffusive systems: it took energy independent value, close to 1/3 (≈ 0.35). While 2 minimum of conductivity was found to significantly overcome the theoretical 4e . πh Second group from Helsinki University (see ref. [35]) have also conducted their study over an important set of samples and succeeded to measure in one of these 2 samples the minimal conductivity close to 4e as well as the predicted Fano factor πh variation (although, the quantitative agreement was not reached). As for the other samples they measured, the results are less obvious to interpret. Hence, based on these two works (some of the results are shown in fig. 13), no indisputable conclusion about the validity of the theory in ref. [148] can be drawn and a third study in this direction could shine light on this question. In the Nanoelectronics group of SPEC (CEA Saclay) where my PhD work took place under direction of professor D. C. Glattli, all type of noise measurements is the key skill. Furthermore, a project of Terahertz detector based on the measurement of Photon-Assisted shot noise in a graphene sample, placed in an antenna is taking off and an experience in noise measurements in graphene was necessary. Accordingly, a real necessity of an extra experiment to once again test the above-mentioned theoretical result together with the availability in the group of the recognized competences to realize such an experiment and finally interest in this topic in view of applications for Terahertz detection are motivated the present work. On the other hand, for this same reason, i.e. because the midterm goal of this 2. quantity characterizing current fluctuations.

(29) CONTENTS. 5. [42] DiCarlo et al. (2008) [35] Danneau et al. (2008). Figure 13: Previous Results. Measurement by DiCarlo et al. of resistance (a) and Fano factor (b). From [42]. (c) and (d) — measurement of the same quantities by Danneau et al., from [35]. project is precisely to attempt to settle up the disagreement between two already performed experiments, the project is particularly challenging. Namely, this meant that it was necessary to conceive the samples and the system for noise measurements, taking into account all limitations of above-mentioned studies and, above all, manage to attain better experimental conditions and thus obtain sufficiently reliable data to conclude about the results of these studies..

(30) 6. CONTENTS. The realization of this project thus dictated the following tasks to be accomplished: • Design and fabricate graphene-based devices for the experiment • Build up the experimental setup • Tune, calibrate and test the measurement system • Investigate the conductance and Fano factor in graphene To fulfil the above-mentioned requirements it was decided to use pristine exfoliated graphene samples and fabricate an experimental device without the back-gate (to avoid the capacitive shunt), replacing it by the side-gates. It would also be perfect to suspend the graphene layer (since it is the best way to reach high mobilities), but this was not accomplished due to the technical difficulties. Measurement approach was also carefully thought over: four-point and cross-correlation techniques were employed, cryogenic low-noise amplifiers together with band-pass filters were used for noise detection. Hence, conductance and Fano factor were successfully investigated in a graphene sample being in the diffusive regime. Technical difficulties together with time constraints prevented us from realizing the same measurements in a ballistic sample, but the obtained results proved the attainability of the ballistic limit in our samples as well as the conformity of the characteristics of our experimental setup to the requirements of the prospected goals. The results of my work are described in the present manuscript. It is organised in the following way: the first part provides the brief insight into both graphene and mesoscopic physics. As for the second part, the details concerning our experimental approach are clarified in one chapter, while another one is devoted to the device fabrication. In the last part our own results are presented..

(31) Part I Overview of Graphene Physics and Mesoscopic Transport. 7.

(32)

(33) Chapter 1 Physical Properties of Graphene 1.1. Crystal Structure of Graphene. Graphene is best known under its graphite form which is simply a stack of graphene planes, held by van der Waals forces and is the most widespread of all carbon allotropes. A single carbon atom has a following electronic structure: 1s2 2s2 2p2 . However, it turns out that, while forming a covalent bonds with other atoms, it is often energetically favourable to promote one electron from 2s orbital to 2p orbital, as a resulting bond energy will be lower in that case. Moreover, one 2s and two 2p orbitals (px and py for example) can hybridize into three sp2 orbitals with 120° angle between them (see: fig. 1.1.a) as it occurs in graphene. These hybridized orbitals will form strong covalent in-plane σ bonds between carbon atoms, that will therefore condense in a hexagonal (sometimes also called honeycomb) crystal lattice with interatomic distance a of 0.142 nm. The remaining pz orbitals point out of plane, forming weaker π bonds. It is the π electrons, delocalized over the whole crystal, that give rise to the valence band states. Crystalline solid is a type of solid in which atoms are arranged with certain periodicity in a way that will entail several translational and rotational symmetries. The common method of representing these symmetries is Bravais lattice, which is an infinite lattice generated by all such vectors, that will leave the crystal, if considered as infinite, invariant under translation by any of these vectors. It usually writes down (in two dimensions) as: R� = l1 a1 + l2 a2. (1.1). with � components l1 and l2 being integer, and a1 and a2 known as primitive vectors of Bravais lattice or basis vectors. These are the smallest possible translations which leave the crystal invariant. Crystal symmetries will play an important role in the band structure calculations, presented in the next section. 9.

(34) 10. CHAPTER 1. PHYSICAL PROPERTIES OF GRAPHENE. . . .   . . . . . . .

(35) . . 

(36) . . . Figure 1.1: (a) Graphene honeycomb lattice with two sublattices highlighted with different colours. Interatomic distance a = 0.142 nm. Also shown primitive vectors of Bravais lattice a1 and a2 as well as vectors δ 1 , δ 2 and δ 3 linking any atom from A sublattice to its three next neighbours. (b) First Brillouin zone (inner hexagon) centred at the point noted Γ, while its corners are noted K and K � (only two are inequivalent). Reciprocal lattice basis vectors a1∗ and a2∗ are also shown. Besides, honeycomb lattice is not a Bravais lattice and the corresponding Bravais lattice will be a triangular1 one with 2 atoms per unit cell, denoted A and B, yielding two sublattices A and B, as represented in fig. 1.1.a. In that case the Bravais lattice basis vectors are: √ √ √ � 3a � a1 = 3aex and a2 = ex + 3ey . (1.2) 2 The following vectors link any atom from A sublattice to its next neighbours, which are necessarily atoms from B sublattice (for any atom from B sublattice it suffices to change the sign): � � a �√ a� √ δ1 = 3ex + ex , δ2 = − 3ex + ex , δ 3 = −aey . (1.3) 2 2 The equivalent of Bravais lattice in the reciprocal space (k-space) is called Reciprocal Lattice and it is no less important in the band structure calculations. As Bravais lattice it is generated in two-dimensional reciprocal space by an ensemble of such vectors: Kλ = λ1 a1∗ + λ2 a2∗ 1. (1.4). In most literature (see [84] for exemple) it is this lattice, that is referred as honeycomb, but we will use the above-mentioned denomination to avoid confusion..

(37) 1.2. GRAPHENE BAND STRUCTURE. 11. that a plane wave of any of these wave vectors Kλ will have the same periodicity as the initial Bravais lattice. This appears in the following conditions: eıKλ R� = 1 for any R� , which constrains λi to be integers and gives the following relation for reciprocal lattice basis vectors aj∗ : ai · aj∗ = 2πδij Another consequence is that two waves of wave vectors that differ from each other by any reciprocal lattice vector Kλ are equivalent. Thus all quantum states of the system are described by the wave vectors inside the First Brillouin Zone (FBZ) — a primitive cell consisting of all points of reciprocal space, that are nearer to the given site of the reciprocal lattice than to all other sites of the lattice. The reciprocal lattice of graphene crystal is spanned by the following vectors: a1∗. �. 2π ey =√ ex − √ 3a 3. �. and. a2∗ =. 4π ey . 3a. (1.5). The FBZ is represented in fig. 1.1.b. As we will see in the next section, the lowenergy excitations wave vectors belong to the vicinity of the two particular points K and K � , situated on the corners of FBZ and represented by the vectors: 4π ±K = ± √ ex . 3 3a. 1.2. (1.6). Graphene Band Structure. In this section we will discuss the electronic properties of graphene, obtained within the tight-binding approximation. As it was mentioned previously it is π electrons, that constitute the valence band, therefore the discussion will be reduced to these states only. In particular, we will consider low-energy excitations, as they are relevant for instance for mesoscopic transport description.. 1.2.1. Basic Principles. Tight-binding model is one of the most basic approaches to electronic band structure calculation, that carries a certain number of approximations. It is however sufficiently powerful to describe in the case of graphene the most prominent aspects of its electronic properties. This approach considers a solid as a collection of almost isolated, only very few interacting atoms. In such a vision it is rather natural to take the orbital states of a single isolated atom as a basis for construction of the electronic states of the solid. The simplest approximation is to consider only highest occupied (valence).

(38) 12. CHAPTER 1. PHYSICAL PROPERTIES OF GRAPHENE. level of the atom Ev and its wavefunction χv (r), assuming that in the solid the energies Ek of the eigenstates ϕk (r) are situated in the very vicinity of Ev , but far from Ev−1 and Ee — first excited state. Consider a single particle Hamiltonian for the R� lattice site: N� 1 ,N2 p2� + Vat (r� − Rm ) H� = 2m0 m1 =1. (1.7). m2 =1. with N1 × N2 — total number of atoms in the crystal. In some cases, we will also use another representation of the single particle Hamiltonian in which it is p2 decomposed in sum of a single atomic Hamiltonian H�at = 2m�0 + Vat (r� − R� ) and the residual potential term ∆V = H� − H�at . It is then reasonable to write trial wavefunction for these eigenstates as a Linear Combination of Atomic Orbitals 2 : ϕk (r) = √. � 1 ak,m χv (r − Rm ) . N1 N2 m. (1.8). As it was discussed in the previous section, crystal lattice exhibits several translational symmetries. Obviously, the trial wavefunctions have to respect these symmetries. These considerations are formalized by Bloch’s Theorem. This very important theorem of the solid state physics exploits the following facts. Because a translation by any Bravais lattice vector R� will leave the physical problem invariant, the �corresponding translational operator TR� should commute with the � Hamiltonian, TR� , H = 0. Hence the eigenstates of latter are necessarily also eigenstates of TR� . In particular, Bloch’s Theorem stipulates that these eigenstates are so called Bloch waves, functions that can be written as a product of a plane wave and a periodic function of the same periodicity as the Bravais lattice. One can easily check, that the following trial wavefunction fulfills the above-mentioned requirement, i.e. it is an eigenstate of the translation operator TR� and moreover, it can be decomposed in a product of a plane wave and a periodic function: ϕk (r) = √. � 1 eıkRm χv (r − Rm ) . N1 N2 m. (1.9). So the electronic band structure calculations in the next section will be based on these trial wavefunctions. 2. The sum. �. m. is a short notation for. �N1 ,N2. m1 ,m2 ..

(39) 1.2. GRAPHENE BAND STRUCTURE. 1.2.2. 13. Band Structure Calculations. We now apply the above-presented model to the case of the graphene honeycomb lattice. As it was already mentioned, lattice symmetry is reflected by Bravais lattice, while honeycomb lattice can only be generated by a Bravais lattice with two atoms per unit cell, or in other words by two sublattices each having the same translational symmetry as the initial lattice. Thus a trial wavefunction, that exhibits the lattice symmetry should in general be described by a linear combination of Bloch waves of two sublattices: B ψk (r) = ak ϕA k (r) + bk ϕk (r − δ AB ). (1.10). Hψk = �k ψk .. (1.11). Here, we have chosen the A sublattice sites to coincide with the sites of Bravais lattice, whereas the B sublattice sites are shifted by δ AB with respect to them, hence this component appearance in the second function’s argument. Both functions B ϕA k (r) and ϕk (r) are given by (1.9), as the atoms of two sublattices are identical, the A and B indices are put here for pedagogical reason, we will omit them from now on. We may now search the solutions of the Schr¨odinger equation with the Hamiltonian H of the form (1.7), but where the ions of both sublattices contribute to the potential energy. Given that there are two electrons per lattice site (one from each atom A and B), multiplication of eq. 1.11 by ψk† on the left leads to a 2 × 2 matrices secular equation: det [Hk − �k Sk ] = 0. with Hkij. =. �. where δ =. 2. dr �. ϕ†k (r)Hϕk (r. 0 −δ AB. δ AB 0. �. − δ ) and ij. Skij. =. �. (1.12). d2 r ϕ†k (r)ϕk (r − δij ),. (1.13). and H and S are the Hamiltonian and overlap matrices. respectively. In the following, in order to simplify the calculations, we will do two approximations. First, we will neglect the overlaps between atomic orbitals, meaning that S is just the identity matrix, intention which is quite reasonable, since pz orbitals have rather limited spatial extension in the xy plane as compared to interatomic distance a. Second, since tunnelling probability diminishes exponentially with the distance, we will restrain the hopping integrals to the nearest-neighbours, meaning that �. d2 r χ† (r − R� )∆V χ(r ± δ AB − Rm ) �� 0. (1.14).

(40) 14. CHAPTER 1. PHYSICAL PROPERTIES OF GRAPHENE. if and only if R� ± δ AB − Rm = δ 1 , δ 2 or δ 3 from (1.3). Moreover, from symmetry considerations the value of these integrals must be � and m independent.3 This value is called hopping amplitude and we denote it as t: t≡. �. d2 r χ† (r)∆V χ(r ± δ AB ).. (1.15). We also introduce for convenience the following constant: t0 ≡. �. d2 r χ† (r)∆V χ(r).. (1.16). As it can be seen from fig. 1.1.a, any A atom has three next neighbours: B1 , B2 , B3 , though only one B atom is described by the same Bravais lattice vector R� as the A atom. The other two B atoms correspond to the lattice vector shifted, in notations accepted in fig. 1.1.a, by vectors a2 and a3 ≡ a2 − a1 . Therefore, they contribute a phase factors exp (ık · a2 ) and exp (ık · a3 ) respectively. So the contribution from a single A�atom to the off-diagonal upper (lower) Hamiltonian matrix (†) element will be4 tγk N1 N2 , where we have defined the sum of next neighbours phase factors: γk ≡ 1 + e−ık·a2 + e−ık·a3. (1.17). Finally, denoting as ε0 the eigenvalue of the atomic Hamiltonian H at , corresponding to χ(r) orbital, we obtain the secular equation in the following form: . with two solutions:. . ε0 + t0 − εk t · γk  det  =0 † t · γk ε0 + t0 − εk. ε± k = ±t|γk | =. � � 3 � � � ±t 3 + 2 cos (k · a. i). (1.18). (1.19). i=1. and we have omitted ε0 + t0 constant, the energy band shift, introduced by it being physically irrelevant. The energy dispersion, calculated in the approximation 3. To verify, whether these approximations are justified, one can perform numerical calculations of the corrections from the further extension of the hopping integrals to the next-nearestneighbours: tnnn , as well as of the corrections from nearest-neighbours orbitals overlap: snn . Such calculations, done in [126] as well as tight-binding fit to cyclotron resonance experiments in [39] show that tnnn − snn t � 0.1t (with t ≈ −3 eV), approving the admitted approximations. 4 N1 × N2 , being the number of A or B atom, and not the total number of atoms as in the previous section..

(41) 1.2. GRAPHENE BAND STRUCTURE. 15. [158] Williams (2009) [26] Castro Neto et al. (2009). Figure 1.2: (a) Graphene energy dispersion obtained within the tight-binding approximation neglecting the next-nearest-neighbours hopping (i.e. tnnn = 0). Right: Dirac cone — energy band near the K and K � (Dirac) points. From [158].(b) Density of states per unit cell as a function of energy (in units of t) also in the assumption of tnnn = 0. From [26]. that takes into account the nearest-neighbour hopping only (tnnn = 0) is plotted in fig. 1.2.a. One distinguishes two bands, that touches each other in several points (though only two of them lay in FBZ and are inequivalent). Lower band + called valence band and upper — conductance band — correspond to ε− k and εk respectively. Since there are as many states as electrons, furnished by carbon atoms, but each state can be occupied by two electrons (with a spin-up and a spin-down) the valence band is completely filled, whereas conductance band is empty. This means, that the Fermi layer passes through the points of contact of two bands, which are called Dirac points for reasons, that will become clear in the next section. From (1.19) it is rather obvious that they correspond to wave-vectors kD , which satisfy: − ε+ k D = εk D = 0. (1.20). relation, that is itself satisfied only if ReγkD = ImγkD = 0, that leads to solutions: 4π kD = ±K = ± √ ex 3 3a. (1.21). in which we recognize two crystallographic points K and K � situated on the FBZ corners, already mentioned in sec. 1.1. Note that due to the time-reversal symmetry, (ε−k = εk ), there are necessarily two inequivalent Dirac points. Therefore, zero-energy states, and by continuity all low-energy states will be doubly degenerate, This degeneracy is referred as valley degeneracy. In the following we will often restrain the discussion to one of the valleys only, but the extension of the results to the remaining valley is straight-forward..

(42) 16. 1.2.3. CHAPTER 1. PHYSICAL PROPERTIES OF GRAPHENE. Low-Energy Excitations. For the following discussion it can be useful to rewrite the problem in slightly different notations. We first introduce effective tight-binding Hamiltonian: . . 0 γk†  Hk ≡ t  γk 0. (1.22). The eigenstates of this Hamiltonian are 2-components vectors, or spinors: Ψ± k. �. a± k = ± bk. �. (1.23). ± The two components a± k and bk of these spinors coincide with the probability amplitudes of the electron to be found on A or B sublattices respectively (the ± indicator stands for valence and conductance bands). These eigenstates are easily derivable, taking into consideration the above-found eigenvalues of (1.22):. Ψ± k where we have defined the angle. �. �. 1 1 =√ , ıφ 2 ±e k. (1.24). �. (1.25). Imγk φk = arctan Reγk. �. As expected, since both sublattices are composed of the identical atoms, the eigenstates correspond to an equal probability for an electron to be found on A and B sublattices. We now consider the low energy excitations of the system, that we define as all states with characteristic energy with respect to the Fermi level being much smaller than the band width. Its wave-vectors therefore lay in the vicinity of the Dirac points, so one can expand the dispersion relation around K (−K case reasonings are analogous) as following: let k = K + q and we take q, that satisfy |q| � |K| ∼ 1/a, so finally the expansion small parameter will be |q|a � 1 and we will limit it to the first order in |q|a. The only k dependent parameter of the effective Hamiltonian and energy dispersion being γk we first expand it: γ�q ≡ γk=K+q =1 + eıK·a2 eıq·a2 + eıK·a3 eıq·a3 �. �. �1 + eı2π/3 1 + ıq · a2 + O(|q|2 a2 ) �. �. + eı2π/3 1 + ıq · a3 + O(|q|2 a2 ) 3a = − (qx + ıqy ) 2. (1.26).

(43) 1.2. GRAPHENE BAND STRUCTURE. 17. here we also took advantage of the fact, that by definition γ�0 ≡ γK = 0, removing all q independent terms. As a consequence, the linearised effective low-energy Hamiltonian will be given by: � = �v qσ H q F. (1.27). with vF — Fermi velocity defined as:. vF = −. 3ta 2�. (1.28). and σ ≡ (σ x , σ y ) — Pauli matrices vector (σ � ≡ (σ x , −σ y ) in the case of K � valley): �. �. 0 1 σ = , 1 0 x. and. �. �. 0 −ı σ = . ı 0 y. (1.29). The energy dispersion, therefore, reads (for both valleys): ε± q = ±�vF |q|,. (1.30). which means that the double degeneracy indeed persists also for the low-energy excitations. This result also confirms the equivalence between |q|a � 1 and |εq | � t � conditions, since |εq | = 3ta|q| 2 � t.. A very useful quantity for thermodynamical and transport properties calculations is the Density of States, which allows to switch from discrete summation over all states to integration over energy spectra. Thus D(ε) dε equals to the number of states with the energy in the interval between ε and dε. We will calculate it in the low-energy limit, considering both valleys, so in all we have to append a factor of 4 to our calculations to take spin and valley degeneracies into account. The common approach to the density of states calculations is to identify in the thermodynamical limit: q(ε) (N1 a · N2 a) � �4× dq 2 (2π) q | εq �ε. �. (1.31). 0. �. �. since q vectors in quasi-continuum are spanned by 2π N1 a and 2π N1 a in x and y directions. Then5 : � 5. � � 1 � 2πq 2|ε| D(ε) dε = 4 × dq = 4 dq = dε 2 2 (2π) (2π) π(�vF )2. In such a way defined density of states is in fact density of states per unit surface. (1.32).

(44) 18. CHAPTER 1. PHYSICAL PROPERTIES OF GRAPHENE. where we performed a beforehand integration over all q, corresponding to the same energy. One obtains directly: D(ε) =. 2|ε| π(�vF )2. (1.33). Note that the density of states vanishes linearly at zero energy, this fact has an important impact on transport properties as it will be seen in the Chapter 3. The full density of states, extended to higher energies is represented in fig. 1.2.b. One notice, in particular, the presence of van-Hove singularities due to the saddle points of the energy dispersion at the M points of FBZ (see fig. 1.2.a and fig. 1.1.a). For further details on electronic properties of graphene the reader is referred to [26].. 1.3. Properties of Dirac Fermions. The form of the linearised effective Hamiltonian (1.27) strongly reminds the Hamiltonian from the Dirac Equation — well known in particle physics relativistic wave equation for spin-1⁄2 particle, formulated by Paul Dirac in 1928[43]6 H = cpσ + mσ z. (1.34). if we assume the mass m being equal to zero (here c is the speed of light). This equation predicts in particular the existence of a new kind of particle — positron, indeed observed for the first time by Carl Anderson in 1932[6] . This similarity by the way explains the appellation “Dirac Points”. Also, it is rather common in graphene community to call π-electrons in graphene massless pseudo-relativistic fermions or simply Dirac Fermions. Note, that this Hamiltonian acts on the already mentioned 2-components vectors or spinors. In the particle physics, the direction of such vector reflects the direction of particle spin in the basis of spin up and down states, while in the case of graphene we will call the related observable pseudo-spin (to distinguish it from its “proper” counterpart), and the up and down states will correspond to two sublattices A and B. In the following we will reveal several curious properties of Dirac equation, that do not occur in the case of ordinary, non-relativistic Schr¨odinger equation. 6. This is a two-dimensional version of the equation.

(45) 1.3. PROPERTIES OF DIRAC FERMIONS. 1.3.1. 19. Probabilty Current Density. In non-relativistic quantum mechanics the probability current density is calculated as follows: j(r, t) =. � � � † ψ ∇ψ − ψ∇ψ † . 2ım. (1.35). In the relativistic case this quantity is derived in the same manner. From Dirac equation, making use of Pauli matrices hermicity, we have ∂Ψ† = −c∇Ψ† σ ∂t. ∂Ψ = −cσ∇Ψ ∂t. (1.36). then the time derivative of probability density is given by ∂ ∂ρ ∂Ψ ∂Ψ† = (Ψ† Ψ) = Ψ† + Ψ ∂t ∂t � ∂t ∂t � = − c Ψ† σ · ∇Ψ + ∇Ψ† · σΨ �. (1.37). �. = − ∇ · cΨ† σΨ .. Since from continuity equation. ∂ρ = −∇ · j(r, t) ∂t. (1.38). j = cΨ† σΨ. (1.39). we finally obtain. a result, that is different from (1.35).. 1.3.2. Zitterbewegung, Chirality and Klein Tunneling. Another intriguing consequence of Dirac equation, named by Erwin Schr¨odinger with German term Zitterbewegung [133] , in other words jittery or trembling motion, states that any attempt to localize a relativistic quantum particle will fail, since it is inevitably accompanied by creation of particle-antiparticle pairs at the position of localization. Indeed, the momentum uncertainty of a confined relativistic particle impacts its energy uncertainty, in contrast to the non-relativistic case, where position-momentum and energy-time uncertainties are unrelated. As a consequence, such confined particle will be described by a wave-packet containing both particle (positive energy) and antiparticle (negative energy) components, which moreover will interfere, causing rapid oscillations of the particle. It is hence.

(46) 20. CHAPTER 1. PHYSICAL PROPERTIES OF GRAPHENE. [82] Katsnelson et al. (2006). Figure 1.3: (a) Sketch illustrating Klein tunnelling through potential barrier. On top: energy bands filling (blue) in different regions, energy branches of the same color are related to the same spin direction. (b) Transmission probability T through a 100 nm-wide barrier as a function of the incident angle for barrier heights of 200 meV (red curve) and 280 meV (Fermi energy of incident electrons ≈ 80 meV. From [82]. impossible to measure the coordinates of a relativistic particle. In a condensed matter system, this phenomena is manifested by the means of special kind of inter-band transitions with creation of virtual electron-hole pairs. Before considering the phenomenon of Klein Tunneling, it is expedient to first introduce one important property of Dirac fermions, namely Chirality or (helicity), defined as the projection of its spin onto the direction of propagation. The corresponding operator reads p·σ hp = (1.40) p The eigenvalues of this operator are ±1, and moreover it obviously commute with the Hamiltonial meaning that the chirality is a constant of motion. The particle-like eigenstates have eigenvalue +1, while anti-particle-like states have eigenvalue −1 (in graphene this is the case for K valley, for K � value it is the other way round). Dirac particles are thus called chiral. This singular property of Dirac fermions leads to what is known as Klein Paradox [86] , a phenomenon which can be resumed as the absence of backscattering from a potential barrier. Consider for simplicity a Dirac particle in a one-dimensional space, which hits a potential barrier (see fig. 1.3.a). One would naturally suppose the particle to be scattered back by the barrier. Yet this would mean that the particle have changed.

(47) 1.3. PROPERTIES OF DIRAC FERMIONS. 21. its propagation direction to the opposite without changing its spin, hence inverting the chirality. This is however impossible, since the latter is the constant of motion. As a consequence, such particle penetrates the barrier. Two-dimensional case is a bit more complicated, the transmission probability depends on the incidence angle, as it can be seen in fig. 1.3.b. This effect was experimentally observed in graphene independently by Young and Kim and by Stander et al..

(48) 22. CHAPTER 1. PHYSICAL PROPERTIES OF GRAPHENE.

Figure

Figure 4: R´esultats pr´ec´edents. Mesure par DiCarlo et al. de la r´esistance (a) et du facteur de Fano (b)
Figure 7: La r´esistance de la constriction en fonction de la tension de grille (vert et bleu)
Figure 8: Un graphique couleur de la r´esistance mesur´ee avec la combinaison crois´ee des sondes de tension (voir le texte) en fonction de ∆V G = V G − V D et du champ magn´etique B
Figure 13: Previous Results. Measurement by DiCarlo et al. of resistance (a) and Fano factor (b)
+7

Références

Documents relatifs

The current fluctuation spectrum at a single tunneling tip is determined by local partial densities of states.. The current-correlation spectrum between two tunneling tips is

We investigate the transition from sub-Poissonian to super-Poissonian values of the zero-temperature shot noise power of a resonant double barrier of macroscopic cross-section..

We investigate the effect of incoherent scattering in a Hanbury Brown and Twiss situation with electrons in edge states of a three-terminal conductor submitted to a strong

We thus see that even though inelastic scattering modi- fies both the conductance and shot noise of the resonant tunneling structure, it leaves the Fano factor unchanged..

Shot noise of series quantum point contacts forming a sequence of cavities in a two-dimensional electron gas are studied theoretically and experimentally.. Noise in such a

1 The decrease of the noise temperature with increasing inelastic rate can be understood as follows: a sufficiently large voltage probe acts as an absorber of rare e − h pairs

We show how, by a suitable complete set of measurements, the elements of the density matrices of the one- and two-particle states can be directly expressed in terms of the currents

Despite the gap, σ(0) originates only from inter-band transitions between the ZEM impurity band and the valence/conduction bands. conduction) band, and for a small but finite